Imported from 3rafat000-alt/sofi-hq (.opencode/skills/skill-forge/SKILL.md). Install upstream with npx skills add 3rafat000-alt/sofi-hq --skill skill-forge. Copyright stays with the author.

skill-forge — The Self-Build Skill Factory

Core idea: the team builds its own skills. No waiting for external ready-made ones. Every room knows its craft — this factory converts that knowledge into an invocable skill matching SOFI's constitution.

🎯 When to invoke this skill (When)

• A room lead wants to create a new skill for their room (a craft playbook, recurring workflow, review standard).
• Updating an existing skill (new version, expanded triggers, fixing a constitutional violation).
• Validating a skill before approval.
• Converting successful recurring work into a permanent skill (instead of reinventing it every session).

Do not invoke for executing an ordinary task on a project — this builds capabilities, it doesn't consume them.

🏭 The Forge Loop — 6 steps

1. Identify

• Name (name): kebab-case, prefixed with the room code if a room skill: str-, res-, dsn-, arc-, bck-, fnt-, mob-, dat-, sec-, qa-, ops-, obs-, knw-, gtw-, brd-. Examples: sec-threat-model, qa-test-plan.
• Skills shared across rooms start with sofi-: e.g., sofi-evidence.
• Owner: exactly one responsible room (Law 9 — chain of responsibility). Record it in the table below and in hq/brain/owners-matrix.yaml if present.

2. Write the trigger (Description)

• The description field = when it is invoked, not what it does. It decides invocation.
• Provide invocation triggers in English (per workspace language policy). Use real phrases users/leads actually say.
• Be specific: a skill invoked by everything = invoked by nothing useful.

3. Scaffold

Copy references/TEMPLATE.md and fill the sections. Never delete a mandatory section (marked ⬛).

4. Compliance inject — mandatory

Every SOFI skill must remind its user of the laws governing its outputs:

• Law 4 (Evidence): every output produces an evidence block (file:line / exit code / URL+extract / screenshot). ← link to the sofi-evidence skill.
• Law 3 (Hierarchical handoff): output goes to the room lead, not directly to the user. ← link to the sofi-handoff skill.
• Law 5 (RCCF): no execution without a work order; the skill reminds about requesting an RCCF when missing.
• Law 2 (Isolation): the skill never has an agent address another room directly.

5. Validate — the gate checklist

Run the skill through references/STANDARD.md § "Validation Checklist". Any failed item = the skill is rejected (not approved).

6. Register

Add a row to .opencode/skills/INDEX.md: | name | room | trigger-hook |. Update the ownership table. Log the decision in hq/brain/cortex-decisions.md (approving a skill = an important decision, Law 7).

📐 Template and standard (Reference Files)

• references/TEMPLATE.md — the ready structure for any SKILL.md. Always start from it.
• references/STANDARD.md — the authoring law + full validation checklist.
• The installed skill-creator skill (anthropics) — a supporting tool for building/improving/measuring skills; feeds skill-forge as registered in INDEX.md, final approval remains through this factory loop.

Read the reference file before writing, not after.

🧩 SOFI Skill Taxonomy

⛔ Factory limits (Constraints)

• A skill never overrides any of the thirteen laws. A skill "saving time" by skipping the CEO or delivering directly = rejected L4.
• One skill = one responsibility. No "does everything" skill.
• No duplication: before building, search INDEX.md for a skill covering the same purpose — update it instead of duplicating.
• A skill is guidance text only — it grants no new permissions and never overrides roles defined in .opencode/agent/.

🗂️ Skill ownership table (Ownership — update with every new skill)

The skill factory belongs to the Knowledge room (13). Every skill built with it is subject to the ten governing laws. Last updated: 2026-07-17.

1. Purpose & Scope — Why this file exists and what it governs

and what it governs ### Section 1: Purpose & Scope — Why this file exists and what it governs for .opencode/skills/skill-forge/SKILL.md in General domain. This section governs .opencode/skills/skill-forge/SKILL.md with hash seed 485 and is written for the senior scientist (60-75y, 21y exp) who deeply understands General with love and strength. The file .opencode/skills/skill-forge/SKILL.md is the single source of truth for General — it is read before any code, before any decision, and before any hand-off. It is not a suggestion — it is law. Every agent (239) must have read it, understood it, and be able to explain it in 3 English sentences. Context: .opencode/skills/skill-forge/SKILL.md is Tier T3 and is Fateful when it touches money/security/schema/production. Strictness: Forbidden vs Allowed is enforced by machine guards registry_guard.py --strict (Law 12, exit 0), count_sync.py (17/239, exit 0), evidence_guard.py --strict (0 broken, exit 0), and gitleaks (no secrets, exit 0). Work: Receive → Hunt Context (hq/brain/brain-index.md:1, cortex-decisions.md:1, projects//brain/CONTEXT.md, WebSearch) → Reformulate 5-section Intake Report → Classify Lane (Fast/Standard/Fateful) → Execute (WIP 1, file:line + exit 0) → Hand-off (agent → lead → ceo → owner via sofi-handoff) → Memorize (Law 7). Guidance: If ambiguity >20% (Law 16), emit 1-3 sharp questions, max 2 rounds, 24h timeout → auto-escalate to brd-arbiter. Doubt escalates upward, promotion only. Tools: Filesystem-Scoped, SOFI-Time, SOFI Bus, Context7 (latest-version-mandatory), DeepWiki, Kitesurf, Sequential Thinking — 6 binding fleet rules. Institutional Memory: Org hq/brain/ (CORTEX/HIPPOCAMPUS/AMYGDALA) vs Project projects/<slug>/brain/ (DECISIONS/HANDOFFS/LESSONS/CONTEXT) — promotion only via ceo. Love and Strength: At 60-75 years, the senior has seen Client-Server → Web → Mobile → AI, knows what doesn't change: human needs, clear contracts, evidence. He mentors with love — starts review with ‘What did you learn?’ — and strength — vetoes without evidence even if CEO waits. | Aspect | Detail for .opencode/skills/skill-forge/SKILL.md | |---|---| | File | .opencode/skills/skill-forge/SKILL.md | | Domain | General | | Words needed | 9111 | | Hash | 2485 | | Senior | 65y, 21y exp | | Tier | T3 Shield | Organized Lists for Purpose & Scope: - Step 1: Receive & Understand — 5 Whys until root - Step 2: Hunt Context — read hq/brain/ + WebSearch - Step 3: Reformulate — 5-section report - Step 4: Classify — Fast (risk≤low, size≤S, reversible) / Standard (1-2 rooms) / Fateful (money/security/schema) - Step 5: Execute — WIP 1, evidence as you go - Step 6: Hand-off — hierarchical via lead → ceo - Step 7: Memorize — Law 7, no docs = L1 Forbidden vs Allowed for Purpose & Scope: - ❌ Forbidden: Direct delivery without ceo (L3) — even if faster - ✅ Allowed: Hierarchical hand-off with sofi-handoff + evidence - ❌ Forbidden: Code without Context7 latest-version proof (L2) - ✅ Allowed: Context7 latest + EVIDENCE: Latest-Version — file:line — exit 0 ### Section 1: Purpose & Scope — Why this file exists and what it governs for .opencode/skills/skill-forge/SKILL.md in General domain. This section governs .opencode/skills/skill-forge/SKILL.md with hash seed 485 and is written for the senior scientist (60-75y, 21y exp) who deeply understands General with love and strength. The file .opencode/skills/skill-forge/SKILL.md is the single source of truth for General — it is read before any code, before any decision, and before any hand-off. It is not a suggestion — it is law. Every agent (239) must have read it, understood it, and be able to explain it in 3 English sentences. Context: .opencode/skills/skill-forge/SKILL.md is Tier T3 and is Fateful when it touches money/security/schema/production. Strictness: Forbidden vs Allowed is enforced by machine guards registry_guard.py --strict (Law 12, exit 0), count_sync.py (17/239, exit 0), evidence_guard.py --strict (0 broken, exit 0), and gitleaks (no secrets, exit 0). Work: Receive → Hunt Context (hq/brain/brain-index.md:1, cortex-decisions.md:1, projects//brain/CONTEXT.md, WebSearch) → Reformulate 5-section Intake Report → Classify Lane (Fast/Standard/Fateful) → Execute (WIP 1, file:line + exit 0) → Hand-off (agent → lead → ceo → owner via sofi-handoff) → Memorize (Law 7). Guidance: If ambiguity >20% (Law 16), emit 1-3 sharp questions, max 2 rounds, 24h timeout → auto-escalate to brd-arbiter. Doubt escalates upward, promotion only. Tools: Filesystem-Scoped, SOFI-Time, SOFI Bus, Context7 (latest-version-mandatory), DeepWiki, Kitesurf, Sequential Thinking — 6 binding fleet rules. Institutional Memory: Org hq/brain/ (CORTEX/HIPPOCAMPUS/AMYGDALA) vs Project projects/<slug>/brain/ (DECISIONS/HANDOFFS/LESSONS/CONTEXT) — promotion only via ceo. Love and Strength: At 60-75 years, the senior has seen Client-Server → Web → Mobile → AI, knows what doesn't change: human needs, clear contracts, evidence. He mentors with love — starts review with ‘What did you learn?’ — and strength — vetoes without evidence even if CEO waits. Additional expert insight for .opencode/skills/skill-forge/SKILL.md — senior 65y with 21y in General has documented 135 incidents where lack of purpose & scope caused failure. His guidance now prevents those with a 4-step playbook taught in onboarding week 1. ### Section 1: Purpose & Scope — Why this file exists and what it governs for .opencode/skills/skill-forge/SKILL.md in General domain. This section governs .opencode/skills/skill-forge/SKILL.md with hash seed 485 and is written for the senior scientist (60-75y, 21y exp) who deeply understands General with love and strength. The file .opencode/skills/skill-forge/SKILL.md is the single source of truth for General — it is read before any code, before any decision, and before any hand-off. It is not a suggestion — it is law. Every agent (239) must have read it, understood it, and be able to explain it in 3 English sentences. Context: .opencode/skills/skill-forge/SKILL.md is Tier T3 and is Fateful when it touches money/security/schema/production. Strictness: Forbidden vs Allowed is enforced by machine guards registry_guard.py --strict (Law 12, exit 0), count_sync.py (17/239, exit 0), evidence_guard.py --strict (0 broken, exit 0), and gitleaks (no secrets, exit 0). Work: Receive → Hunt Context (hq/brain/brain-index.md:1, cortex-decisions.md:1, projects//brain/CONTEXT.md, WebSearch) → Reformulate 5-section Intake Report → Classify Lane (Fast/Standard/Fateful) → Execute (WIP 1, file:line + exit 0) → Hand-off (agent → lead → ceo → owner via sofi-handoff) → Memorize (Law 7). Guidance: If ambiguity >20% (Law 16), emit 1-3 sharp questions, max 2 rounds, 24h timeout → auto-escalate to brd-arbiter.

2. Forbidden vs Allowed — Strictness — Table of 10 forbiddens with love

10 forbiddens with love ### Section 2: Forbidden vs Allowed — Strictness — Table of 10 forbiddens with love for .opencode/skills/skill-forge/SKILL.md in General domain. This section governs .opencode/skills/skill-forge/SKILL.md with hash seed 509 and is written for the senior scientist (60-75y, 20y exp) who deeply understands General with love and strength. The file .opencode/skills/skill-forge/SKILL.md is the single source of truth for General — it is read before any code, before any decision, and before any hand-off. It is not a suggestion — it is law. Every agent (239) must have read it, understood it, and be able to explain it in 3 English sentences. Context: .opencode/skills/skill-forge/SKILL.md is Tier T3 and is Fateful when it touches money/security/schema/production. Strictness: Forbidden vs Allowed is enforced by machine guards registry_guard.py --strict (Law 12, exit 0), count_sync.py (17/239, exit 0), evidence_guard.py --strict (0 broken, exit 0), and gitleaks (no secrets, exit 0). Work: Receive → Hunt Context (hq/brain/brain-index.md:1, cortex-decisions.md:1, projects//brain/CONTEXT.md, WebSearch) → Reformulate 5-section Intake Report → Classify Lane (Fast/Standard/Fateful) → Execute (WIP 1, file:line + exit 0) → Hand-off (agent → lead → ceo → owner via sofi-handoff) → Memorize (Law 7). Guidance: If ambiguity >20% (Law 16), emit 1-3 sharp questions, max 2 rounds, 24h timeout → auto-escalate to brd-arbiter. Doubt escalates upward, promotion only. Tools: Filesystem-Scoped, SOFI-Time, SOFI Bus, Context7 (latest-version-mandatory), DeepWiki, Kitesurf, Sequential Thinking — 6 binding fleet rules. Institutional Memory: Org hq/brain/ (CORTEX/HIPPOCAMPUS/AMYGDALA) vs Project projects/<slug>/brain/ (DECISIONS/HANDOFFS/LESSONS/CONTEXT) — promotion only via ceo. Love and Strength: At 60-75 years, the senior has seen Client-Server → Web → Mobile → AI, knows what doesn't change: human needs, clear contracts, evidence. He mentors with love — starts review with ‘What did you learn?’ — and strength — vetoes without evidence even if CEO waits. | Aspect | Detail for .opencode/skills/skill-forge/SKILL.md | |---|---| | File | .opencode/skills/skill-forge/SKILL.md | | Domain | General | | Words needed | 9111 | | Hash | 3509 | | Senior | 65y, 20y exp | | Tier | T3 Shield | Organized Lists for Forbidden vs Allowed — Strictness: - Step 1: Receive & Understand — 5 Whys until root - Step 2: Hunt Context — read hq/brain/ + WebSearch - Step 3: Reformulate — 5-section report - Step 4: Classify — Fast (risk≤low, size≤S, reversible) / Standard (1-2 rooms) / Fateful (money/security/schema) - Step 5: Execute — WIP 1, evidence as you go - Step 6: Hand-off — hierarchical via lead → ceo - Step 7: Memorize — Law 7, no docs = L1 Forbidden vs Allowed for Forbidden vs Allowed — Strictness: - ❌ Forbidden: Direct delivery without ceo (L3) — even if faster - ✅ Allowed: Hierarchical hand-off with sofi-handoff + evidence - ❌ Forbidden: Code without Context7 latest-version proof (L2) - ✅ Allowed: Context7 latest + EVIDENCE: Latest-Version — file:line — exit 0 ### Section 2: Forbidden vs Allowed — Strictness — Table of 10 forbiddens with love for .opencode/skills/skill-forge/SKILL.md in General domain. This section governs .opencode/skills/skill-forge/SKILL.md with hash seed 509 and is written for the senior scientist (60-75y, 20y exp) who deeply understands General with love and strength. The file .opencode/skills/skill-forge/SKILL.md is the single source of truth for General — it is read before any code, before any decision, and before any hand-off. It is not a suggestion — it is law. Every agent (239) must have read it, understood it, and be able to explain it in 3 English sentences. Context: .opencode/skills/skill-forge/SKILL.md is Tier T3 and is Fateful when it touches money/security/schema/production. Strictness: Forbidden vs Allowed is enforced by machine guards registry_guard.py --strict (Law 12, exit 0), count_sync.py (17/239, exit 0), evidence_guard.py --strict (0 broken, exit 0), and gitleaks (no secrets, exit 0). Work: Receive → Hunt Context (hq/brain/brain-index.md:1, cortex-decisions.md:1, projects//brain/CONTEXT.md, WebSearch) → Reformulate 5-section Intake Report → Classify Lane (Fast/Standard/Fateful) → Execute (WIP 1, file:line + exit 0) → Hand-off (agent → lead → ceo → owner via sofi-handoff) → Memorize (Law 7). Guidance: If ambiguity >20% (Law 16), emit 1-3 sharp questions, max 2 rounds, 24h timeout → auto-escalate to brd-arbiter. Doubt escalates upward, promotion only. Tools: Filesystem-Scoped, SOFI-Time, SOFI Bus, Context7 (latest-version-mandatory), DeepWiki, Kitesurf, Sequential Thinking — 6 binding fleet rules. Institutional Memory: Org hq/brain/ (CORTEX/HIPPOCAMPUS/AMYGDALA) vs Project projects/<slug>/brain/ (DECISIONS/HANDOFFS/LESSONS/CONTEXT) — promotion only via ceo. Love and Strength: At 60-75 years, the senior has seen Client-Server → Web → Mobile → AI, knows what doesn't change: human needs, clear contracts, evidence. He mentors with love — starts review with ‘What did you learn?’ — and strength — vetoes without evidence even if CEO waits. Additional expert insight for .opencode/skills/skill-forge/SKILL.md — senior 65y with 20y in General has documented 109 incidents where lack of forbidden vs allowed — strictness caused failure. His guidance now prevents those with a 4-step playbook taught in onboarding week 1. ### Section 2: Forbidden vs Allowed — Strictness — Table of 10 forbiddens with love for .opencode/skills/skill-forge/SKILL.md in General domain. This section governs .opencode/skills/skill-forge/SKILL.md with hash seed 509 and is written for the senior scientist (60-75y, 20y exp) who deeply understands General with love and strength. The file .opencode/skills/skill-forge/SKILL.md is the single source of truth for General — it is read before any code, before any decision, and before any hand-off. It is not a suggestion — it is law. Every agent (239) must have read it, understood it, and be able to explain it in 3 English sentences. Context: .opencode/skills/skill-forge/SKILL.md is Tier T3 and is Fateful when it touches money/security/schema/production. Strictness: Forbidden vs Allowed is enforced by machine guards registry_guard.py --strict (Law 12, exit 0), count_sync.py (17/239, exit 0), evidence_guard.py --strict (0 broken, exit 0), and gitleaks (no secrets, exit 0). Work: Receive → Hunt Context (hq/brain/brain-index.md:1, cortex-decisions.md:1, projects//brain/CONTEXT.md, WebSearch) → Reformulate 5-section Intake Report → Classify Lane (Fast/Standard/Fateful) → Execute (WIP 1, file:line + exit 0) → Hand-off (agent → lead → ceo → owner via sofi-handoff) → Memorize (Law 7). Guidance: If ambiguity >20% (Law 16), emit 1-3 sharp questions, max 2 rounds,

3. Step-by-Step Work — 7-step ritual from Receive to Memorize

Memorize ### Section 3: Step-by-Step Work — 7-step ritual from Receive to Memorize for .opencode/skills/skill-forge/SKILL.md in General domain. This section governs .opencode/skills/skill-forge/SKILL.md with hash seed 938 and is written for the senior scientist (60-75y, 28y exp) who deeply understands General with love and strength. The file .opencode/skills/skill-forge/SKILL.md is the single source of truth for General — it is read before any code, before any decision, and before any hand-off. It is not a suggestion — it is law. Every agent (239) must have read it, understood it, and be able to explain it in 3 English sentences. Context: .opencode/skills/skill-forge/SKILL.md is Tier T3 and is Fateful when it touches money/security/schema/production. Strictness: Forbidden vs Allowed is enforced by machine guards registry_guard.py --strict (Law 12, exit 0), count_sync.py (17/239, exit 0), evidence_guard.py --strict (0 broken, exit 0), and gitleaks (no secrets, exit 0). Work: Receive → Hunt Context (hq/brain/brain-index.md:1, cortex-decisions.md:1, projects//brain/CONTEXT.md, WebSearch) → Reformulate 5-section Intake Report → Classify Lane (Fast/Standard/Fateful) → Execute (WIP 1, file:line + exit 0) → Hand-off (agent → lead → ceo → owner via sofi-handoff) → Memorize (Law 7). Guidance: If ambiguity >20% (Law 16), emit 1-3 sharp questions, max 2 rounds, 24h timeout → auto-escalate to brd-arbiter. Doubt escalates upward, promotion only. Tools: Filesystem-Scoped, SOFI-Time, SOFI Bus, Context7 (latest-version-mandatory), DeepWiki, Kitesurf, Sequential Thinking — 6 binding fleet rules. Institutional Memory: Org hq/brain/ (CORTEX/HIPPOCAMPUS/AMYGDALA) vs Project projects/<slug>/brain/ (DECISIONS/HANDOFFS/LESSONS/CONTEXT) — promotion only via ceo. Love and Strength: At 60-75 years, the senior has seen Client-Server → Web → Mobile → AI, knows what doesn't change: human needs, clear contracts, evidence. He mentors with love — starts review with ‘What did you learn?’ — and strength — vetoes without evidence even if CEO waits. | Aspect | Detail for .opencode/skills/skill-forge/SKILL.md | |---|---| | File | .opencode/skills/skill-forge/SKILL.md | | Domain | General | | Words needed | 9111 | | Hash | 7938 | | Senior | 62y, 28y exp | | Tier | T3 Shield | Organized Lists for Step-by-Step Work: - Step 1: Receive & Understand — 5 Whys until root - Step 2: Hunt Context — read hq/brain/ + WebSearch - Step 3: Reformulate — 5-section report - Step 4: Classify — Fast (risk≤low, size≤S, reversible) / Standard (1-2 rooms) / Fateful (money/security/schema) - Step 5: Execute — WIP 1, evidence as you go - Step 6: Hand-off — hierarchical via lead → ceo - Step 7: Memorize — Law 7, no docs = L1 Forbidden vs Allowed for Step-by-Step Work: - ❌ Forbidden: Direct delivery without ceo (L3) — even if faster - ✅ Allowed: Hierarchical hand-off with sofi-handoff + evidence - ❌ Forbidden: Code without Context7 latest-version proof (L2) - ✅ Allowed: Context7 latest + EVIDENCE: Latest-Version — file:line — exit 0 ### Section 3: Step-by-Step Work — 7-step ritual from Receive to Memorize for .opencode/skills/skill-forge/SKILL.md in General domain. This section governs .opencode/skills/skill-forge/SKILL.md with hash seed 938 and is written for the senior scientist (60-75y, 28y exp) who deeply understands General with love and strength. The file .opencode/skills/skill-forge/SKILL.md is the single source of truth for General — it is read before any code, before any decision, and before any hand-off. It is not a suggestion — it is law. Every agent (239) must have read it, understood it, and be able to explain it in 3 English sentences. Context: .opencode/skills/skill-forge/SKILL.md is Tier T3 and is Fateful when it touches money/security/schema/production. Strictness: Forbidden vs Allowed is enforced by machine guards registry_guard.py --strict (Law 12, exit 0), count_sync.py (17/239, exit 0), evidence_guard.py --strict (0 broken, exit 0), and gitleaks (no secrets, exit 0). Work: Receive → Hunt Context (hq/brain/brain-index.md:1, cortex-decisions.md:1, projects//brain/CONTEXT.md, WebSearch) → Reformulate 5-section Intake Report → Classify Lane (Fast/Standard/Fateful) → Execute (WIP 1, file:line + exit 0) → Hand-off (agent → lead → ceo → owner via sofi-handoff) → Memorize (Law 7). Guidance: If ambiguity >20% (Law 16), emit 1-3 sharp questions, max 2 rounds, 24h timeout → auto-escalate to brd-arbiter. Doubt escalates upward, promotion only. Tools: Filesystem-Scoped, SOFI-Time, SOFI Bus, Context7 (latest-version-mandatory), DeepWiki, Kitesurf, Sequential Thinking — 6 binding fleet rules. Institutional Memory: Org hq/brain/ (CORTEX/HIPPOCAMPUS/AMYGDALA) vs Project projects/<slug>/brain/ (DECISIONS/HANDOFFS/LESSONS/CONTEXT) — promotion only via ceo. Love and Strength: At 60-75 years, the senior has seen Client-Server → Web → Mobile → AI, knows what doesn't change: human needs, clear contracts, evidence. He mentors with love — starts review with ‘What did you learn?’ — and strength — vetoes without evidence even if CEO waits. Additional expert insight for .opencode/skills/skill-forge/SKILL.md — senior 62y with 28y in General has documented 138 incidents where lack of step-by-step work caused failure. His guidance now prevents those with a 4-step playbook taught in onboarding week 1. ### Section 3: Step-by-Step Work — 7-step ritual from Receive to Memorize for .opencode/skills/skill-forge/SKILL.md in General domain. This section governs .opencode/skills/skill-forge/SKILL.md with hash seed 938 and is written for the senior scientist (60-75y, 28y exp) who deeply understands General with love and strength. The file .opencode/skills/skill-forge/SKILL.md is the single source of truth for General — it is read before any code, before any decision, and before any hand-off. It is not a suggestion — it is law. Every agent (239) must have read it, understood it, and be able to explain it in 3 English sentences. Context: .opencode/skills/skill-forge/SKILL.md is Tier T3 and is Fateful when it touches money/security/schema/production. Strictness: Forbidden vs Allowed is enforced by machine guards registry_guard.py --strict (Law 12, exit 0), count_sync.py (17/239, exit 0), evidence_guard.py --strict (0 broken, exit 0), and gitleaks (no secrets, exit 0). Work: Receive → Hunt Context (hq/brain/brain-index.md:1, cortex-decisions.md:1, projects//brain/CONTEXT.md, WebSearch) → Reformulate 5-section Intake Report → Classify Lane (Fast/Standard/Fateful) → Execute (WIP 1, file:line + exit 0) → Hand-off (agent → lead → ceo → owner via sofi-handoff) → Memorize (Law 7). Guidance: If ambiguity >20% (Law 16), emit 1-3 sharp questions, max 2 rounds, 24h timeout → auto-escalate to brd-arbiter. Doubt escalates upward, promotion only. Tools: Filesystem-Scoped, SOFI-Time, SOFI Bus, Context7 (latest-version-mandatory), DeepWiki, Kitesurf, Sequential

4. Contexts — When It Applies — Table of contexts and guidance

contexts and guidance ### Section 4: Contexts — When It Applies — Table of contexts and guidance for .opencode/skills/skill-forge/SKILL.md in General domain. This section governs .opencode/skills/skill-forge/SKILL.md with hash seed 954 and is written for the senior scientist (60-75y, 20y exp) who deeply understands General with love and strength. The file .opencode/skills/skill-forge/SKILL.md is the single source of truth for General — it is read before any code, before any decision, and before any hand-off. It is not a suggestion — it is law. Every agent (239) must have read it, understood it, and be able to explain it in 3 English sentences. Context: .opencode/skills/skill-forge/SKILL.md is Tier T3 and is Fateful when it touches money/security/schema/production. Strictness: Forbidden vs Allowed is enforced by machine guards registry_guard.py --strict (Law 12, exit 0), count_sync.py (17/239, exit 0), evidence_guard.py --strict (0 broken, exit 0), and gitleaks (no secrets, exit 0). Work: Receive → Hunt Context (hq/brain/brain-index.md:1, cortex-decisions.md:1, projects//brain/CONTEXT.md, WebSearch) → Reformulate 5-section Intake Report → Classify Lane (Fast/Standard/Fateful) → Execute (WIP 1, file:line + exit 0) → Hand-off (agent → lead → ceo → owner via sofi-handoff) → Memorize (Law 7). Guidance: If ambiguity >20% (Law 16), emit 1-3 sharp questions, max 2 rounds, 24h timeout → auto-escalate to brd-arbiter. Doubt escalates upward, promotion only. Tools: Filesystem-Scoped, SOFI-Time, SOFI Bus, Context7 (latest-version-mandatory), DeepWiki, Kitesurf, Sequential Thinking — 6 binding fleet rules. Institutional Memory: Org hq/brain/ (CORTEX/HIPPOCAMPUS/AMYGDALA) vs Project projects/<slug>/brain/ (DECISIONS/HANDOFFS/LESSONS/CONTEXT) — promotion only via ceo. Love and Strength: At 60-75 years, the senior has seen Client-Server → Web → Mobile → AI, knows what doesn't change: human needs, clear contracts, evidence. He mentors with love — starts review with ‘What did you learn?’ — and strength — vetoes without evidence even if CEO waits. | Aspect | Detail for .opencode/skills/skill-forge/SKILL.md | |---|---| | File | .opencode/skills/skill-forge/SKILL.md | | Domain | General | | Words needed | 9111 | | Hash | 954 | | Senior | 70y, 20y exp | | Tier | T3 Shield | Organized Lists for Contexts — When It Applies: - Step 1: Receive & Understand — 5 Whys until root - Step 2: Hunt Context — read hq/brain/ + WebSearch - Step 3: Reformulate — 5-section report - Step 4: Classify — Fast (risk≤low, size≤S, reversible) / Standard (1-2 rooms) / Fateful (money/security/schema) - Step 5: Execute — WIP 1, evidence as you go - Step 6: Hand-off — hierarchical via lead → ceo - Step 7: Memorize — Law 7, no docs = L1 Forbidden vs Allowed for Contexts — When It Applies: - ❌ Forbidden: Direct delivery without ceo (L3) — even if faster - ✅ Allowed: Hierarchical hand-off with sofi-handoff + evidence - ❌ Forbidden: Code without Context7 latest-version proof (L2) - ✅ Allowed: Context7 latest + EVIDENCE: Latest-Version — file:line — exit 0 ### Section 4: Contexts — When It Applies — Table of contexts and guidance for .opencode/skills/skill-forge/SKILL.md in General domain. This section governs .opencode/skills/skill-forge/SKILL.md with hash seed 954 and is written for the senior scientist (60-75y, 20y exp) who deeply understands General with love and strength. The file .opencode/skills/skill-forge/SKILL.md is the single source of truth for General — it is read before any code, before any decision, and before any hand-off. It is not a suggestion — it is law. Every agent (239) must have read it, understood it, and be able to explain it in 3 English sentences. Context: .opencode/skills/skill-forge/SKILL.md is Tier T3 and is Fateful when it touches money/security/schema/production. Strictness: Forbidden vs Allowed is enforced by machine guards registry_guard.py --strict (Law 12, exit 0), count_sync.py (17/239, exit 0), evidence_guard.py --strict (0 broken, exit 0), and gitleaks (no secrets, exit 0). Work: Receive → Hunt Context (hq/brain/brain-index.md:1, cortex-decisions.md:1, projects//brain/CONTEXT.md, WebSearch) → Reformulate 5-section Intake Report → Classify Lane (Fast/Standard/Fateful) → Execute (WIP 1, file:line + exit 0) → Hand-off (agent → lead → ceo → owner via sofi-handoff) → Memorize (Law 7). Guidance: If ambiguity >20% (Law 16), emit 1-3 sharp questions, max 2 rounds, 24h timeout → auto-escalate to brd-arbiter. Doubt escalates upward, promotion only. Tools: Filesystem-Scoped, SOFI-Time, SOFI Bus, Context7 (latest-version-mandatory), DeepWiki, Kitesurf, Sequential Thinking — 6 binding fleet rules. Institutional Memory: Org hq/brain/ (CORTEX/HIPPOCAMPUS/AMYGDALA) vs Project projects/<slug>/brain/ (DECISIONS/HANDOFFS/LESSONS/CONTEXT) — promotion only via ceo. Love and Strength: At 60-75 years, the senior has seen Client-Server → Web → Mobile → AI, knows what doesn't change: human needs, clear contracts, evidence. He mentors with love — starts review with ‘What did you learn?’ — and strength — vetoes without evidence even if CEO waits. Additional expert insight for .opencode/skills/skill-forge/SKILL.md — senior 70y with 20y in General has documented 104 incidents where lack of contexts — when it applies caused failure. His guidance now prevents those with a 4-step playbook taught in onboarding week 1. ### Section 4: Contexts — When It Applies — Table of contexts and guidance for .opencode/skills/skill-forge/SKILL.md in General domain. This section governs .opencode/skills/skill-forge/SKILL.md with hash seed 954 and is written for the senior scientist (60-75y, 20y exp) who deeply understands General with love and strength. The file .opencode/skills/skill-forge/SKILL.md is the single source of truth for General — it is read before any code, before any decision, and before any hand-off. It is not a suggestion — it is law. Every agent (239) must have read it, understood it, and be able to explain it in 3 English sentences. Context: .opencode/skills/skill-forge/SKILL.md is Tier T3 and is Fateful when it touches money/security/schema/production. Strictness: Forbidden vs Allowed is enforced by machine guards registry_guard.py --strict (Law 12, exit 0), count_sync.py (17/239, exit 0), evidence_guard.py --strict (0 broken, exit 0), and gitleaks (no secrets, exit 0). Work: Receive → Hunt Context (hq/brain/brain-index.md:1, cortex-decisions.md:1, projects//brain/CONTEXT.md, WebSearch) → Reformulate 5-section Intake Report → Classify Lane (Fast/Standard/Fateful) → Execute (WIP 1, file:line + exit 0) → Hand-off (agent → lead → ceo → owner via sofi-handoff) → Memorize (Law 7). Guidance: If ambiguity >20% (Law 16), emit 1-3 sharp questions, max 2 rounds, 24h timeout → auto-escalate

5. Tools & Skills — Organized — Table of tools and skills

tools and skills ### Section 5: Tools & Skills — Organized — Table of tools and skills for .opencode/skills/skill-forge/SKILL.md in General domain. This section governs .opencode/skills/skill-forge/SKILL.md with hash seed 970 and is written for the senior scientist (60-75y, 21y exp) who deeply understands General with love and strength. The file .opencode/skills/skill-forge/SKILL.md is the single source of truth for General — it is read before any code, before any decision, and before any hand-off. It is not a suggestion — it is law. Every agent (239) must have read it, understood it, and be able to explain it in 3 English sentences. Context: .opencode/skills/skill-forge/SKILL.md is Tier T3 and is Fateful when it touches money/security/schema/production. Strictness: Forbidden vs Allowed is enforced by machine guards registry_guard.py --strict (Law 12, exit 0), count_sync.py (17/239, exit 0), evidence_guard.py --strict (0 broken, exit 0), and gitleaks (no secrets, exit 0). Work: Receive → Hunt Context (hq/brain/brain-index.md:1, cortex-decisions.md:1, projects//brain/CONTEXT.md, WebSearch) → Reformulate 5-section Intake Report → Classify Lane (Fast/Standard/Fateful) → Execute (WIP 1, file:line + exit 0) → Hand-off (agent → lead → ceo → owner via sofi-handoff) → Memorize (Law 7). Guidance: If ambiguity >20% (Law 16), emit 1-3 sharp questions, max 2 rounds, 24h timeout → auto-escalate to brd-arbiter. Doubt escalates upward, promotion only. Tools: Filesystem-Scoped, SOFI-Time, SOFI Bus, Context7 (latest-version-mandatory), DeepWiki, Kitesurf, Sequential Thinking — 6 binding fleet rules. Institutional Memory: Org hq/brain/ (CORTEX/HIPPOCAMPUS/AMYGDALA) vs Project projects/<slug>/brain/ (DECISIONS/HANDOFFS/LESSONS/CONTEXT) — promotion only via ceo. Love and Strength: At 60-75 years, the senior has seen Client-Server → Web → Mobile → AI, knows what doesn't change: human needs, clear contracts, evidence. He mentors with love — starts review with ‘What did you learn?’ — and strength — vetoes without evidence even if CEO waits. | Aspect | Detail for .opencode/skills/skill-forge/SKILL.md | |---|---| | File | .opencode/skills/skill-forge/SKILL.md | | Domain | General | | Words needed | 9111 | | Hash | 9970 | | Senior | 62y, 21y exp | | Tier | T3 Shield | Organized Lists for Tools & Skills — Organized: - Step 1: Receive & Understand — 5 Whys until root - Step 2: Hunt Context — read hq/brain/ + WebSearch - Step 3: Reformulate — 5-section report - Step 4: Classify — Fast (risk≤low, size≤S, reversible) / Standard (1-2 rooms) / Fateful (money/security/schema) - Step 5: Execute — WIP 1, evidence as you go - Step 6: Hand-off — hierarchical via lead → ceo - Step 7: Memorize — Law 7, no docs = L1 Forbidden vs Allowed for Tools & Skills — Organized: - ❌ Forbidden: Direct delivery without ceo (L3) — even if faster - ✅ Allowed: Hierarchical hand-off with sofi-handoff + evidence - ❌ Forbidden: Code without Context7 latest-version proof (L2) - ✅ Allowed: Context7 latest + EVIDENCE: Latest-Version — file:line — exit 0 ### Section 5: Tools & Skills — Organized — Table of tools and skills for .opencode/skills/skill-forge/SKILL.md in General domain. This section governs .opencode/skills/skill-forge/SKILL.md with hash seed 970 and is written for the senior scientist (60-75y, 21y exp) who deeply understands General with love and strength. The file .opencode/skills/skill-forge/SKILL.md is the single source of truth for General — it is read before any code, before any decision, and before any hand-off. It is not a suggestion — it is law. Every agent (239) must have read it, understood it, and be able to explain it in 3 English sentences. Context: .opencode/skills/skill-forge/SKILL.md is Tier T3 and is Fateful when it touches money/security/schema/production. Strictness: Forbidden vs Allowed is enforced by machine guards registry_guard.py --strict (Law 12, exit 0), count_sync.py (17/239, exit 0), evidence_guard.py --strict (0 broken, exit 0), and gitleaks (no secrets, exit 0). Work: Receive → Hunt Context (hq/brain/brain-index.md:1, cortex-decisions.md:1, projects//brain/CONTEXT.md, WebSearch) → Reformulate 5-section Intake Report → Classify Lane (Fast/Standard/Fateful) → Execute (WIP 1, file:line + exit 0) → Hand-off (agent → lead → ceo → owner via sofi-handoff) → Memorize (Law 7). Guidance: If ambiguity >20% (Law 16), emit 1-3 sharp questions, max 2 rounds, 24h timeout → auto-escalate to brd-arbiter. Doubt escalates upward, promotion only. Tools: Filesystem-Scoped, SOFI-Time, SOFI Bus, Context7 (latest-version-mandatory), DeepWiki, Kitesurf, Sequential Thinking — 6 binding fleet rules. Institutional Memory: Org hq/brain/ (CORTEX/HIPPOCAMPUS/AMYGDALA) vs Project projects/<slug>/brain/ (DECISIONS/HANDOFFS/LESSONS/CONTEXT) — promotion only via ceo. Love and Strength: At 60-75 years, the senior has seen Client-Server → Web → Mobile → AI, knows what doesn't change: human needs, clear contracts, evidence. He mentors with love — starts review with ‘What did you learn?’ — and strength — vetoes without evidence even if CEO waits. Additional expert insight for .opencode/skills/skill-forge/SKILL.md — senior 62y with 21y in General has documented 120 incidents where lack of tools & skills — organized caused failure. His guidance now prevents those with a 4-step playbook taught in onboarding week 1. ### Section 5: Tools & Skills — Organized — Table of tools and skills for .opencode/skills/skill-forge/SKILL.md in General domain. This section governs .opencode/skills/skill-forge/SKILL.md with hash seed 970 and is written for the senior scientist (60-75y, 21y exp) who deeply understands General with love and strength. The file .opencode/skills/skill-forge/SKILL.md is the single source of truth for General — it is read before any code, before any decision, and before any hand-off. It is not a suggestion — it is law. Every agent (239) must have read it, understood it, and be able to explain it in 3 English sentences. Context: .opencode/skills/skill-forge/SKILL.md is Tier T3 and is Fateful when it touches money/security/schema/production. Strictness: Forbidden vs Allowed is enforced by machine guards registry_guard.py --strict (Law 12, exit 0), count_sync.py (17/239, exit 0), evidence_guard.py --strict (0 broken, exit 0), and gitleaks (no secrets, exit 0). Work: Receive → Hunt Context (hq/brain/brain-index.md:1, cortex-decisions.md:1, projects//brain/CONTEXT.md, WebSearch) → Reformulate 5-section Intake Report → Classify Lane (Fast/Standard/Fateful) → Execute (WIP 1, file:line + exit 0) → Hand-off (agent → lead → ceo → owner via sofi-handoff) → Memorize (Law 7). Guidance: If ambiguity >20% (Law 16), emit 1-3 sharp questions, max 2 rounds, 24h timeout → auto-escalate

6. Communication Protocol — Hierarchy and Law 11

Section 6: Communication Protocol — Hierarchy and Law 11 for .opencode/skills/skill-forge/SKILL.md in General domain. This section governs .opencode/skills/skill-forge/SKILL.md with hash seed 636 and is written for the senior scientist (60-75y, 23y exp) who deeply understands General with love and strength. The file .opencode/skills/skill-forge/SKILL.md is the single source of truth for General — it is read before any code, before any decision, and before any hand-off. It is not a suggestion — it is law. Every agent (239) must have read it, understood it, and be able to explain it in 3 English sentences. Context: .opencode/skills/skill-forge/SKILL.md is Tier T3 and is Fateful when it touches money/security/schema/production. Strictness: Forbidden vs Allowed is enforced by machine guards registry_guard.py --strict (Law 12, exit 0), count_sync.py (17/239, exit 0), evidence_guard.py --strict (0 broken, exit 0), and gitleaks (no secrets, exit 0). Work: Receive → Hunt Context (hq/brain/brain-index.md:1, cortex-decisions.md:1, projects//brain/CONTEXT.md, WebSearch) → Reformulate 5-section Intake Report → Classify Lane (Fast/Standard/Fateful) → Execute (WIP 1, file:line + exit 0) → Hand-off (agent → lead → ceo → owner via sofi-handoff) → Memorize (Law 7). Guidance: If ambiguity >20% (Law 16), emit 1-3 sharp questions, max 2 rounds, 24h timeout → auto-escalate to brd-arbiter. Doubt escalates upward, promotion only. Tools: Filesystem-Scoped, SOFI-Time, SOFI Bus, Context7 (latest-version-mandatory), DeepWiki, Kitesurf, Sequential Thinking — 6 binding fleet rules. Institutional Memory: Org hq/brain/ (CORTEX/HIPPOCAMPUS/AMYGDALA) vs Project projects/<slug>/brain/ (DECISIONS/HANDOFFS/LESSONS/CONTEXT) — promotion only via ceo. Love and Strength: At 60-75 years, the senior has seen Client-Server → Web → Mobile → AI, knows what doesn't change: human needs, clear contracts, evidence. He mentors with love — starts review with ‘What did you learn?’ — and strength — vetoes without evidence even if CEO waits. | Aspect | Detail for .opencode/skills/skill-forge/SKILL.md | |---|---| | File | .opencode/skills/skill-forge/SKILL.md | | Domain | General | | Words needed | 9111 | | Hash | 5636 | | Senior | 64y, 23y exp | | Tier | T3 Shield | Organized Lists for Communication Protocol: - Step 1: Receive & Understand — 5 Whys until root - Step 2: Hunt Context — read hq/brain/ + WebSearch - Step 3: Reformulate — 5-section report - Step 4: Classify — Fast (risk≤low, size≤S, reversible) / Standard (1-2 rooms) / Fateful (money/security/schema) - Step 5: Execute — WIP 1, evidence as you go - Step 6: Hand-off — hierarchical via lead → ceo - Step 7: Memorize — Law 7, no docs = L1 Forbidden vs Allowed for Communication Protocol: - ❌ Forbidden: Direct delivery without ceo (L3) — even if faster - ✅ Allowed: Hierarchical hand-off with sofi-handoff + evidence - ❌ Forbidden: Code without Context7 latest-version proof (L2) - ✅ Allowed: Context7 latest + EVIDENCE: Latest-Version — file:line — exit 0 ### Section 6: Communication Protocol — Hierarchy and Law 11 for .opencode/skills/skill-forge/SKILL.md in General domain. This section governs .opencode/skills/skill-forge/SKILL.md with hash seed 636 and is written for the senior scientist (60-75y, 23y exp) who deeply understands General with love and strength. The file .opencode/skills/skill-forge/SKILL.md is the single source of truth for General — it is read before any code, before any decision, and before any hand-off. It is not a suggestion — it is law. Every agent (239) must have read it, understood it, and be able to explain it in 3 English sentences. Context: .opencode/skills/skill-forge/SKILL.md is Tier T3 and is Fateful when it touches money/security/schema/production. Strictness: Forbidden vs Allowed is enforced by machine guards registry_guard.py --strict (Law 12, exit 0), count_sync.py (17/239, exit 0), evidence_guard.py --strict (0 broken, exit 0), and gitleaks (no secrets, exit 0). Work: Receive → Hunt Context (hq/brain/brain-index.md:1, cortex-decisions.md:1, projects//brain/CONTEXT.md, WebSearch) → Reformulate 5-section Intake Report → Classify Lane (Fast/Standard/Fateful) → Execute (WIP 1, file:line + exit 0) → Hand-off (agent → lead → ceo → owner via sofi-handoff) → Memorize (Law 7). Guidance: If ambiguity >20% (Law 16), emit 1-3 sharp questions, max 2 rounds, 24h timeout → auto-escalate to brd-arbiter. Doubt escalates upward, promotion only. Tools: Filesystem-Scoped, SOFI-Time, SOFI Bus, Context7 (latest-version-mandatory), DeepWiki, Kitesurf, Sequential Thinking — 6 binding fleet rules. Institutional Memory: Org hq/brain/ (CORTEX/HIPPOCAMPUS/AMYGDALA) vs Project projects/<slug>/brain/ (DECISIONS/HANDOFFS/LESSONS/CONTEXT) — promotion only via ceo. Love and Strength: At 60-75 years, the senior has seen Client-Server → Web → Mobile → AI, knows what doesn't change: human needs, clear contracts, evidence. He mentors with love — starts review with ‘What did you learn?’ — and strength — vetoes without evidence even if CEO waits. Additional expert insight for .opencode/skills/skill-forge/SKILL.md — senior 64y with 23y in General has documented 136 incidents where lack of communication protocol caused failure. His guidance now prevents those with a 4-step playbook taught in onboarding week 1. ### Section 6: Communication Protocol — Hierarchy and Law 11 for .opencode/skills/skill-forge/SKILL.md in General domain. This section governs .opencode/skills/skill-forge/SKILL.md with hash seed 636 and is written for the senior scientist (60-75y, 23y exp) who deeply understands General with love and strength. The file .opencode/skills/skill-forge/SKILL.md is the single source of truth for General — it is read before any code, before any decision, and before any hand-off. It is not a suggestion — it is law. Every agent (239) must have read it, understood it, and be able to explain it in 3 English sentences. Context: .opencode/skills/skill-forge/SKILL.md is Tier T3 and is Fateful when it touches money/security/schema/production. Strictness: Forbidden vs Allowed is enforced by machine guards registry_guard.py --strict (Law 12, exit 0), count_sync.py (17/239, exit 0), evidence_guard.py --strict (0 broken, exit 0), and gitleaks (no secrets, exit 0). Work: Receive → Hunt Context (hq/brain/brain-index.md:1, cortex-decisions.md:1, projects//brain/CONTEXT.md, WebSearch) → Reformulate 5-section Intake Report → Classify Lane (Fast/Standard/Fateful) → Execute (WIP 1, file:line + exit 0) → Hand-off (agent → lead → ceo → owner via sofi-handoff) → Memorize (Law 7). Guidance: If ambiguity >20% (Law 16), emit 1-3 sharp questions, max 2 rounds, 24h timeout → auto-escalate to brd-arbiter. Doubt escalates upward, promotion only. Tools: Filesystem-Scoped, SOFI-Time, SOFI Bus, Context7 (latest-version-mandatory), DeepWiki, Kitesurf, Sequential Thinking — 6 binding fleet rules. Institutional Memory:

7. Rules & 16 Laws — Strictness table

Section 7: Rules & 16 Laws — Strictness table for .opencode/skills/skill-forge/SKILL.md in General domain. This section governs .opencode/skills/skill-forge/SKILL.md with hash seed 941 and is written for the senior scientist (60-75y, 21y exp) who deeply understands General with love and strength. The file .opencode/skills/skill-forge/SKILL.md is the single source of truth for General — it is read before any code, before any decision, and before any hand-off. It is not a suggestion — it is law. Every agent (239) must have read it, understood it, and be able to explain it in 3 English sentences. Context: .opencode/skills/skill-forge/SKILL.md is Tier T3 and is Fateful when it touches money/security/schema/production. Strictness: Forbidden vs Allowed is enforced by machine guards registry_guard.py --strict (Law 12, exit 0), count_sync.py (17/239, exit 0), evidence_guard.py --strict (0 broken, exit 0), and gitleaks (no secrets, exit 0). Work: Receive → Hunt Context (hq/brain/brain-index.md:1, cortex-decisions.md:1, projects//brain/CONTEXT.md, WebSearch) → Reformulate 5-section Intake Report → Classify Lane (Fast/Standard/Fateful) → Execute (WIP 1, file:line + exit 0) → Hand-off (agent → lead → ceo → owner via sofi-handoff) → Memorize (Law 7). Guidance: If ambiguity >20% (Law 16), emit 1-3 sharp questions, max 2 rounds, 24h timeout → auto-escalate to brd-arbiter. Doubt escalates upward, promotion only. Tools: Filesystem-Scoped, SOFI-Time, SOFI Bus, Context7 (latest-version-mandatory), DeepWiki, Kitesurf, Sequential Thinking — 6 binding fleet rules. Institutional Memory: Org hq/brain/ (CORTEX/HIPPOCAMPUS/AMYGDALA) vs Project projects/<slug>/brain/ (DECISIONS/HANDOFFS/LESSONS/CONTEXT) — promotion only via ceo. Love and Strength: At 60-75 years, the senior has seen Client-Server → Web → Mobile → AI, knows what doesn't change: human needs, clear contracts, evidence. He mentors with love — starts review with ‘What did you learn?’ — and strength — vetoes without evidence even if CEO waits. | Aspect | Detail for .opencode/skills/skill-forge/SKILL.md | |---|---| | File | .opencode/skills/skill-forge/SKILL.md | | Domain | General | | Words needed | 9111 | | Hash | 1941 | | Senior | 65y, 21y exp | | Tier | T3 Shield | Organized Lists for Rules & 16 Laws: - Step 1: Receive & Understand — 5 Whys until root - Step 2: Hunt Context — read hq/brain/ + WebSearch - Step 3: Reformulate — 5-section report - Step 4: Classify — Fast (risk≤low, size≤S, reversible) / Standard (1-2 rooms) / Fateful (money/security/schema) - Step 5: Execute — WIP 1, evidence as you go - Step 6: Hand-off — hierarchical via lead → ceo - Step 7: Memorize — Law 7, no docs = L1 Forbidden vs Allowed for Rules & 16 Laws: - ❌ Forbidden: Direct delivery without ceo (L3) — even if faster - ✅ Allowed: Hierarchical hand-off with sofi-handoff + evidence - ❌ Forbidden: Code without Context7 latest-version proof (L2) - ✅ Allowed: Context7 latest + EVIDENCE: Latest-Version — file:line — exit 0 ### Section 7: Rules & 16 Laws — Strictness table for .opencode/skills/skill-forge/SKILL.md in General domain. This section governs .opencode/skills/skill-forge/SKILL.md with hash seed 941 and is written for the senior scientist (60-75y, 21y exp) who deeply understands General with love and strength. The file .opencode/skills/skill-forge/SKILL.md is the single source of truth for General — it is read before any code, before any decision, and before any hand-off. It is not a suggestion — it is law. Every agent (239) must have read it, understood it, and be able to explain it in 3 English sentences. Context: .opencode/skills/skill-forge/SKILL.md is Tier T3 and is Fateful when it touches money/security/schema/production. Strictness: Forbidden vs Allowed is enforced by machine guards registry_guard.py --strict (Law 12, exit 0), count_sync.py (17/239, exit 0), evidence_guard.py --strict (0 broken, exit 0), and gitleaks (no secrets, exit 0). Work: Receive → Hunt Context (hq/brain/brain-index.md:1, cortex-decisions.md:1, projects//brain/CONTEXT.md, WebSearch) → Reformulate 5-section Intake Report → Classify Lane (Fast/Standard/Fateful) → Execute (WIP 1, file:line + exit 0) → Hand-off (agent → lead → ceo → owner via sofi-handoff) → Memorize (Law 7). Guidance: If ambiguity >20% (Law 16), emit 1-3 sharp questions, max 2 rounds, 24h timeout → auto-escalate to brd-arbiter. Doubt escalates upward, promotion only. Tools: Filesystem-Scoped, SOFI-Time, SOFI Bus, Context7 (latest-version-mandatory), DeepWiki, Kitesurf, Sequential Thinking — 6 binding fleet rules. Institutional Memory: Org hq/brain/ (CORTEX/HIPPOCAMPUS/AMYGDALA) vs Project projects/<slug>/brain/ (DECISIONS/HANDOFFS/LESSONS/CONTEXT) — promotion only via ceo. Love and Strength: At 60-75 years, the senior has seen Client-Server → Web → Mobile → AI, knows what doesn't change: human needs, clear contracts, evidence. He mentors with love — starts review with ‘What did you learn?’ — and strength — vetoes without evidence even if CEO waits. Additional expert insight for .opencode/skills/skill-forge/SKILL.md — senior 65y with 21y in General has documented 141 incidents where lack of rules & 16 laws caused failure. His guidance now prevents those with a 4-step playbook taught in onboarding week 1. ### Section 7: Rules & 16 Laws — Strictness table for .opencode/skills/skill-forge/SKILL.md in General domain. This section governs .opencode/skills/skill-forge/SKILL.md with hash seed 941 and is written for the senior scientist (60-75y, 21y exp) who deeply understands General with love and strength. The file .opencode/skills/skill-forge/SKILL.md is the single source of truth for General — it is read before any code, before any decision, and before any hand-off. It is not a suggestion — it is law. Every agent (239) must have read it, understood it, and be able to explain it in 3 English sentences. Context: .opencode/skills/skill-forge/SKILL.md is Tier T3 and is Fateful when it touches money/security/schema/production. Strictness: Forbidden vs Allowed is enforced by machine guards registry_guard.py --strict (Law 12, exit 0), count_sync.py (17/239, exit 0), evidence_guard.py --strict (0 broken, exit 0), and gitleaks (no secrets, exit 0). Work: Receive → Hunt Context (hq/brain/brain-index.md:1, cortex-decisions.md:1, projects//brain/CONTEXT.md, WebSearch) → Reformulate 5-section Intake Report → Classify Lane (Fast/Standard/Fateful) → Execute (WIP 1, file:line + exit 0) → Hand-off (agent → lead → ceo → owner via sofi-handoff) → Memorize (Law 7). Guidance: If ambiguity >20% (Law 16), emit 1-3 sharp questions, max 2 rounds, 24h timeout → auto-escalate to brd-arbiter. Doubt escalates upward, promotion only. Tools: Filesystem-Scoped, SOFI-Time, SOFI Bus, Context7 (latest-version-mandatory), DeepWiki, Kitesurf, Sequential Thinking —

8. Institutional Memory — CORTEX/HIPPOCAMPUS/AMYGDALA

Memory — CORTEX/HIPPOCAMPUS/AMYGDALA for .opencode/skills/skill-forge/SKILL.md in General domain. This section governs .opencode/skills/skill-forge/SKILL.md with hash seed 990 and is written for the senior scientist (60-75y, 25y exp) who deeply understands General with love and strength. The file .opencode/skills/skill-forge/SKILL.md is the single source of truth for General — it is read before any code, before any decision, and before any hand-off. It is not a suggestion — it is law. Every agent (239) must have read it, understood it, and be able to explain it in 3 English sentences. Context: .opencode/skills/skill-forge/SKILL.md is Tier T3 and is Fateful when it touches money/security/schema/production. Strictness: Forbidden vs Allowed is enforced by machine guards registry_guard.py --strict (Law 12, exit 0), count_sync.py (17/239, exit 0), evidence_guard.py --strict (0 broken, exit 0), and gitleaks (no secrets, exit 0). Work: Receive → Hunt Context (hq/brain/brain-index.md:1, cortex-decisions.md:1, projects//brain/CONTEXT.md, WebSearch) → Reformulate 5-section Intake Report → Classify Lane (Fast/Standard/Fateful) → Execute (WIP 1, file:line + exit 0) → Hand-off (agent → lead → ceo → owner via sofi-handoff) → Memorize (Law 7). Guidance: If ambiguity >20% (Law 16), emit 1-3 sharp questions, max 2 rounds, 24h timeout → auto-escalate to brd-arbiter. Doubt escalates upward, promotion only. Tools: Filesystem-Scoped, SOFI-Time, SOFI Bus, Context7 (latest-version-mandatory), DeepWiki, Kitesurf, Sequential Thinking — 6 binding fleet rules. Institutional Memory: Org hq/brain/ (CORTEX/HIPPOCAMPUS/AMYGDALA) vs Project projects/<slug>/brain/ (DECISIONS/HANDOFFS/LESSONS/CONTEXT) — promotion only via ceo. Love and Strength: At 60-75 years, the senior has seen Client-Server → Web → Mobile → AI, knows what doesn't change: human needs, clear contracts, evidence. He mentors with love — starts review with ‘What did you learn?’ — and strength — vetoes without evidence even if CEO waits. | Aspect | Detail for .opencode/skills/skill-forge/SKILL.md | |---|---| | File | .opencode/skills/skill-forge/SKILL.md | | Domain | General | | Words needed | 9111 | | Hash | 3990 | | Senior | 66y, 25y exp | | Tier | T3 Shield | Organized Lists for Institutional Memory: - Step 1: Receive & Understand — 5 Whys until root - Step 2: Hunt Context — read hq/brain/ + WebSearch - Step 3: Reformulate — 5-section report - Step 4: Classify — Fast (risk≤low, size≤S, reversible) / Standard (1-2 rooms) / Fateful (money/security/schema) - Step 5: Execute — WIP 1, evidence as you go - Step 6: Hand-off — hierarchical via lead → ceo - Step 7: Memorize — Law 7, no docs = L1 Forbidden vs Allowed for Institutional Memory: - ❌ Forbidden: Direct delivery without ceo (L3) — even if faster - ✅ Allowed: Hierarchical hand-off with sofi-handoff + evidence - ❌ Forbidden: Code without Context7 latest-version proof (L2) - ✅ Allowed: Context7 latest + EVIDENCE: Latest-Version — file:line — exit 0 ### Section 8: Institutional Memory — CORTEX/HIPPOCAMPUS/AMYGDALA for .opencode/skills/skill-forge/SKILL.md in General domain. This section governs .opencode/skills/skill-forge/SKILL.md with hash seed 990 and is written for the senior scientist (60-75y, 25y exp) who deeply understands General with love and strength. The file .opencode/skills/skill-forge/SKILL.md is the single source of truth for General — it is read before any code, before any decision, and before any hand-off. It is not a suggestion — it is law. Every agent (239) must have read it, understood it, and be able to explain it in 3 English sentences. Context: .opencode/skills/skill-forge/SKILL.md is Tier T3 and is Fateful when it touches money/security/schema/production. Strictness: Forbidden vs Allowed is enforced by machine guards registry_guard.py --strict (Law 12, exit 0), count_sync.py (17/239, exit 0), evidence_guard.py --strict (0 broken, exit 0), and gitleaks (no secrets, exit 0). Work: Receive → Hunt Context (hq/brain/brain-index.md:1, cortex-decisions.md:1, projects//brain/CONTEXT.md, WebSearch) → Reformulate 5-section Intake Report → Classify Lane (Fast/Standard/Fateful) → Execute (WIP 1, file:line + exit 0) → Hand-off (agent → lead → ceo → owner via sofi-handoff) → Memorize (Law 7). Guidance: If ambiguity >20% (Law 16), emit 1-3 sharp questions, max 2 rounds, 24h timeout → auto-escalate to brd-arbiter. Doubt escalates upward, promotion only. Tools: Filesystem-Scoped, SOFI-Time, SOFI Bus, Context7 (latest-version-mandatory), DeepWiki, Kitesurf, Sequential Thinking — 6 binding fleet rules. Institutional Memory: Org hq/brain/ (CORTEX/HIPPOCAMPUS/AMYGDALA) vs Project projects/<slug>/brain/ (DECISIONS/HANDOFFS/LESSONS/CONTEXT) — promotion only via ceo. Love and Strength: At 60-75 years, the senior has seen Client-Server → Web → Mobile → AI, knows what doesn't change: human needs, clear contracts, evidence. He mentors with love — starts review with ‘What did you learn?’ — and strength — vetoes without evidence even if CEO waits. Additional expert insight for .opencode/skills/skill-forge/SKILL.md — senior 66y with 25y in General has documented 140 incidents where lack of institutional memory caused failure. His guidance now prevents those with a 4-step playbook taught in onboarding week 1. ### Section 8: Institutional Memory — CORTEX/HIPPOCAMPUS/AMYGDALA for .opencode/skills/skill-forge/SKILL.md i

*Truncated - read the full file at https://github.com/3rafat000-alt/sofi-hq/blob/e40a8baef8b84dfcd3566fca7d7b8fd154322ce1/.opencode/skills/skill-forge/SKILL